Most bang for 600 USD?

[Mrs Beanbag]

Yes!  AGA-compatible, with DVI-out.  3GHz "68080" (quad core?).  PCiE.

I am considering a project of my own, making an accelerator for A1200 with an ARM CPU running software emulation.  What is the point of that?  Well lots of people are doing it with FPGAs but it would seem cheaper, faster and easier to do it this way as there are software emu's already available, ARM chips come in the GHz range already and are widely available (and the logical next step for desktop PCs in general if you ask me) and affordable.  The software would be in ROM on the card itself so you wouldn't know it was there.  Why not just emulate the whole Amiga on a PC?  Well because then you have to host the emulator in some other OS you have to boot up, ugh.  And it doesn't feel the same.

After that, remake an A1200 motherboard with AGA in an FPGA, no CPU but with the accelerator edge connector, and no 14MHz limit.  (This is why I asked about core licensing elsewhere, but I think I'll try the accelerator first.)

[Mrs Beanbag]

But AAA wasn't going to be completely compatible with AGA anyway, and since it was never released there is no software for it, and graphics hardware has advanced so much since then it hardly seems worth bothering.  However PCs still maintain legacy text mode/VGA for various reasons so there's no reason we can't have AGA compatibility in there, alongside some more established GPU design from the likes of nVidia.  In fact a PCIe slot would be a must anyway.

I like the way the Sega Megadrive kept the Master System's Z80 as a co-processor, and then the Saturn had a 68000 as a co-processor!  Still, I think these days software emulation of the 68k should suffice, since it would still be faster than any real Motorolla chip that was ever put in an Amiga, as long as the emulator is part of the hardware so it's invisible, and new software can use whatever CPU it really is in its native language.  Which might require the OS to pull some tricks but we'll cross that bridge when we come to it.

[Mrs Beanbag]

You mean everyone wouldn't be excited by a new Amiga?

[Mrs Beanbag]

Well I don't know what excites general tech nerds, except after the event, and it usually surprises me.  But the thing that really impressed people about the Amiga when it came out was raytracing.  I've read a paper about quite an impressive real-time raytracer (several frames per second) implemented in an FPGA that ran at alarmingly low clock speeds, so maybe something like that could do the trick.  Texture mapping and shading has served the games industry well so far but it's not "the real deal" and maybe it stops being the most efficient solution at some level of detail.

[Mrs Beanbag]

http://www-wjp.cs.uni-saarland.de/publikationen/SWWPS04.pdf

GPUs, on the other hand, are made for massive parallelization, but aren't necessarily optimal for full on ray-tracing, for a variety of reasons, impressive though they are.

[Mrs Beanbag]

Indeed, and in fact if you look at it one way, the Blitter is a sort of FPGA - the minterms register is essentially a LUT!

[Mrs Beanbag]

How about a drop in replacement for existing accelerator CPUs using fpga or emulation on Arm or Atom processors, it would be useful to others outside the Amiga community using 68K

That's what I said!

In fact I've been conferring with some techies on this, we've decided that since you have to sync down to 14MHz to communicate with the A1200 chipset, a USB connection would be fast enough.  So build a board that connects the trapdoor slot to a USB socket, that allows you to read and write directly to the CPU buses, connect to a Raspberry Pi, and run the 68k emulator there.  This would be an interesting experiment, if nothing else.

[Mrs Beanbag]

Would you explain the need to "sync down to 14 MHz" and how a serial interface can handle that much bidirectional data?  This is just because my degree was not in engineering, electrical or otherwise.

The A1200 motherboard runs at 14MHz no matter what, if you have a CPU on an accelerator card that runs at other than 14MHz you still have to communicate with the motherboard at 14MHz.

USB* can get 480Mbit/s, although with 32 data lines and 24 address lines it might still be a bit of a bottleneck.  But it's really more for proof-of-concept than final design, we have to take things one step at a time.  It's just that producing a custom ARM board is a lot of unnecessary work at this stage, when we already have an ARM board already available at lost cost.  The idea is get the software working on that first, and we can think about a faster interface later, and ultimately a single-board design.

*USB 2.0

[Mrs Beanbag]

The problem is that the fastest speed is when you're transferring alot of data. You're going to have crippling latency ... Any time you spend on using USB is going to be wasted and the results disappointing and not an indication of what you can achieve.

This is true but when doing development you do one bit at a time, you start simple and get it to work first, then you worry about the performance.  I've never produced anything for the expansion port before, just being able to probe the ROM from my PC would be a start.

[Mrs Beanbag]

When worrying about the performance means throwing everything away then you're doing it wrong. Doing development one bit at a time is good, but you aren't doing any bits that are worth doing.

I won't be throwing everything away, just the USB connection, which is easy.  I still need the 68k emulator for ARM with whatever mods necessary (i.e. all memory access below 01x000000 needs to go off-board), and I still need to interface with the expansion slot somehow.  But it's early days yet, I don't have a formal written plan, so I'm still looking at options.  I'm going into the Hacklab tomorrow night anyway, see if anyone there knows how to build ARM boards, so far I've only spoken to a notorious pessimist.

We can make our own PCBs here.  But if I can't make an ARM board I don't know what chance I'd have with an FPGA...?  What would I need one for, anyway?  I only need CPU, RAM, flash ROM and some kind of buffer.

[Mrs Beanbag]

Well, so much for pub conversations.

Any realistic solution requires a low latency and high speed interface. Thus a GPIO directly to the "guest" CPU. Perhaps the GPIO on STM32 (ARM) CPU or Raspberry-PI is capable of that.

"Realistic" is a long-term goal of course, but GPIO seems like a far better idea anyway!  Seems quite easy to use, as well.

[Mrs Beanbag]

Here is what the article says about emulation
 " introduces new boot code and timing challenges. The effort to resolve these challenges can exceed the appropriate  lifecycle extension budget, consuming any profit to be made in extending  the product’s life.In contrast, simply replacing the obsolete  processor part with a new FPGA device that’s fully hardware and software  compatible is usually significantly easier and less expensive."

The article says a lot of things.  It depends on your design goals, of course, and expertise.  For instance it also says this:
"If your design team has little experience using IP cores and FPGAs, then  using a discrete processor chip is likely the better approach."

And "How many more years do you expect to ship the product? The longer this  life, the greater the chance your new processor chip will also become  obsolete, and the more using IP makes sense."  I don't expect ARM chips to go obsolete any time soon.

Anyway since this is a hobby project (at least for now) most of the business-type decisions aren't really important.  BUT I did consider the FPGA option, and rejected it for several reasons, the central one being, I don't intend merely to replace the 680x0 with an exact replica, but rather, to improve on it.

Also, I already have a Raspberry Pi.

(Besides, that article is basically a piece of marketing for their own IP cores...)

[Mrs Beanbag]

YES, sorry, kind of hijacked your thread there.  I'll start my own thread once I've got something to show for it.

To be honest I've no idea what would impress the current market.  When the iPad came out I thought, what is the point of that?  It doesn't seem to solve any technical problem at all.  There's nothing groundbreaking there from a technical point of view, but it comes in a stylish and convenient package.  It's more of a lifestyle solution than a computing solution.

So if your question is, what would sell really well?  I can't answer that.  I know that a board with built-in FPGAs that the programmer could use could achieve some technically impressive feats, whether the average consumer could care less is another matter.  But the rise of the portable computing market has really put ARM on the scene, and I expect desktop PCs to follow suit eventually.  If I could buy a desktop motherboard based on ARM right now, that beat my current AMD 4850e, I'd be happy.

[Mrs Beanbag]

Is the ARM chip same endian as 680x0 or opposite like the x86?

They can be switched to either mode (bi-endian).